Copyright (C) 2005 Moray Allan
These programs are free software; you can redistribute them and/or modify them under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
These programs are distributed in the hope that they will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with these programs; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
- GSL, BLAS, Atlas libraries.
- MIDI::Simple library, available from CPAN as part of MIDI-Perl.
This contains Bach's chorales as downloaded from ftp://i11ftp.ira.uka.de/pub/neuro/dominik/midifiles/bach.zip
./harmonise.sh
Generates the most probable harmonisations under the model.
Creates a chord model 'model-chords' and finds the most probable harmonisation for each test chorale under this model, then creates an ornamentation model 'model-ornamentation' and finds the most probable ornamentation for each of the generated chords, giving final ornamented harmonisations in the directory 'model-ornamentation/viterbi-results'.
Also evaluates the performance of the models, and prints comparative statistics, given as log-probabilities per symbol (sample cross-entropies).
./notatechorale.sh FILENAME
Typesets a chorale data file in music notation, using the lilypond music notation program.
./playchorale.sh FILENAME
Plays back a chorale data file, using the timidity MIDI synthesiser.
./chorale-c/prob MODELNAME evaluate
Trains an HMM using the input data for the named model (e.g. 'chords'), and evaluates how well it fits the data, also comparing performance against some simpler models.
./chorale-c/prob MODELNAME viterbi
Trains an HMM using the input data for the named model, and finds the most probable state sequence for each entry in the test set of original harmonisations.
./chorale-c/prob MODELNAME viterbi-test
Trains an HMM using the input data for the named model, and finds the most probable state sequences for each entry in the test set, using the sequences in 'input-test' rather than 'input'. This is used to find ornamentations for chord sequences generated by a previous HMM, where the test data is not taken directly from Bach's harmonisations.
./chorale-c/prob MODELNAME sample [seed=N]
Trains an HMM using the input data for the trained model, and samples a state sequence for each entry in the test set. An initial seed for the random number generator may be given.